In light of e.g. http://reactos.com:8080/archives/public/ros-dev/2004-November/000658.html I would like to suggest the following header ideas.
Currently MingW has: 1.A set of headers that contain a copy of part of the platform SDK 2.Some extra stuff (like that ntdll.h) 3.A set of headers that contain a copy of part of the windows DDK 4.Mabie again some extra stuff MS doesnt have (I dont know) 5.A set of headers that contain a copy of part of the visual C++ runtime library 6.Some extra stuff (mainly bits required to deal with differences between GCC and Visual C I think)
WINE has: 1.A set of headers that are supposed to contain a copy of part of the platform SDK (although some things have been claimed to be wrong) 2.A set of headers that are supposed to contain a copy of part of the visual C++ runtime library (although again there may be things that are wrong) and 3.Extra definitions for stuff in both header sets microsoft doesnt define in the platform SDK (e.g. internal bits, functions microsoft doesnt document etc)
ReactOS has: w32api (the ROS import of it) include/ddk (the ReactOS version of the windows DDK) include/wine (the WINE windows API headers) include/msvcrt (the MSVCRT headers ROS uses to build and link with msvcrt) include (various misc ReactOS specific headers)
What I suggest for ReactOS (and I think some of this is already being done) is as follows: grab any headers from include/* that are public DDK headers (if there are any) and make sure they are all in include/ddk grab everything from the current WINE headers MingW is missing. Check that its all correct and stuff and add it to the ReactOS import of w32api. (if anything is found to be wrong in the wine headers, it should be fixed) grab anything from include itself and add to w32api as needed Get permission to relicence the stuff and contribute back to the main w32api tree. grab contents of include/msvcrt on ReactOS plus include/msvcrt on WINE and also mingw-runtime. Create one set of header files from the 3 sets such that it matches the microsoft headers as closely as possible. Merge w32api DDK headers with ROS DDK headers and create one set that, again, matches the MS headers.
For w32api/mingw-runtime, the suggestion is that they look at all the "new stuff" generated by the merging of the 3 sets of msvcrt headers plus the ROS/WINE/w32api merger plus the DDK merger and look at acccepting it.
For WINE, I suggest they start using the new "merged" headers if at all possible.
Basicly, the idea is that mingw-runtime would become the "definitive" set of msvcrt.dll headers out there and that ReactOS, MingW and WINE would start using it for any case where you are talking to msvcrt.dll from the public side. both ROS and WINE would then have their own internal msvcrt headers as necessary to build their msvcrt* builds Then, w32api would become the "definitive" set of win32 api headers and would contain as much of the Microsoft Platform SDK as people have legally cloned. All userland components of ReactOS (as well as kernel-side bits that need the stuff in the userland headers e.g. win32k.sys) would use this. Be nice if WINE were to use this too but because of what WINE is and how it is built/used, there may be reasons for the WINE team not to use w32api as the build headers used to build WINE. And, there would be one "definitive" set of DDK headers which would be good enough for building ReactOS with. And for building any kernel-side stuff WINE decides to add (e.g. that stuff to load the device drivers for vaerious copy protection systems that was being worked on). And for anyone using MingW and MingW GCC to build kernel-mode stuff. Anything private or project specific would go into a set of private headers as needed for the different projects and sub-sections (dlls etc) within them Anything exported by microsoft but undocumented would go into a set of "undocumented" headers (which could then be shared between ReactOS and WINE since I dont think w32api would accept anything not documented in MSDN)
Just suggestions btw, if they are good, say so If they are bad, say so (and say why) If there are issues with my idea, say what they are.
Hi Jonathan,
I am already working on a complete header system overhaul that's been approved and very much requested by the other developers.
Thanks for your suggestions. Some of them aren't so right, but the main parts of what you said I'm already doing, but in a more orderly and simplified fashion to guarnatee support with MS PSDK/DDK/IFS.
I invite you to read one of my older blog entries to find out more about it.
Best Regards, Alex Ionescu
Jonathan Wilson wrote:
In light of e.g. http://reactos.com:8080/archives/public/ros-dev/2004-November/000658.html I would like to suggest the following header ideas.
Currently MingW has: 1.A set of headers that contain a copy of part of the platform SDK 2.Some extra stuff (like that ntdll.h) 3.A set of headers that contain a copy of part of the windows DDK 4.Mabie again some extra stuff MS doesnt have (I dont know) 5.A set of headers that contain a copy of part of the visual C++ runtime library 6.Some extra stuff (mainly bits required to deal with differences between GCC and Visual C I think)
WINE has: 1.A set of headers that are supposed to contain a copy of part of the platform SDK (although some things have been claimed to be wrong) 2.A set of headers that are supposed to contain a copy of part of the visual C++ runtime library (although again there may be things that are wrong) and 3.Extra definitions for stuff in both header sets microsoft doesnt define in the platform SDK (e.g. internal bits, functions microsoft doesnt document etc)
ReactOS has: w32api (the ROS import of it) include/ddk (the ReactOS version of the windows DDK) include/wine (the WINE windows API headers) include/msvcrt (the MSVCRT headers ROS uses to build and link with msvcrt) include (various misc ReactOS specific headers)
What I suggest for ReactOS (and I think some of this is already being done) is as follows: grab any headers from include/* that are public DDK headers (if there are any) and make sure they are all in include/ddk grab everything from the current WINE headers MingW is missing. Check that its all correct and stuff and add it to the ReactOS import of w32api. (if anything is found to be wrong in the wine headers, it should be fixed) grab anything from include itself and add to w32api as needed Get permission to relicence the stuff and contribute back to the main w32api tree. grab contents of include/msvcrt on ReactOS plus include/msvcrt on WINE and also mingw-runtime. Create one set of header files from the 3 sets such that it matches the microsoft headers as closely as possible. Merge w32api DDK headers with ROS DDK headers and create one set that, again, matches the MS headers.
For w32api/mingw-runtime, the suggestion is that they look at all the "new stuff" generated by the merging of the 3 sets of msvcrt headers plus the ROS/WINE/w32api merger plus the DDK merger and look at acccepting it.
For WINE, I suggest they start using the new "merged" headers if at all possible.
Basicly, the idea is that mingw-runtime would become the "definitive" set of msvcrt.dll headers out there and that ReactOS, MingW and WINE would start using it for any case where you are talking to msvcrt.dll from the public side. both ROS and WINE would then have their own internal msvcrt headers as necessary to build their msvcrt* builds Then, w32api would become the "definitive" set of win32 api headers and would contain as much of the Microsoft Platform SDK as people have legally cloned. All userland components of ReactOS (as well as kernel-side bits that need the stuff in the userland headers e.g. win32k.sys) would use this. Be nice if WINE were to use this too but because of what WINE is and how it is built/used, there may be reasons for the WINE team not to use w32api as the build headers used to build WINE. And, there would be one "definitive" set of DDK headers which would be good enough for building ReactOS with. And for building any kernel-side stuff WINE decides to add (e.g. that stuff to load the device drivers for vaerious copy protection systems that was being worked on). And for anyone using MingW and MingW GCC to build kernel-mode stuff. Anything private or project specific would go into a set of private headers as needed for the different projects and sub-sections (dlls etc) within them Anything exported by microsoft but undocumented would go into a set of "undocumented" headers (which could then be shared between ReactOS and WINE since I dont think w32api would accept anything not documented in MSDN)
Just suggestions btw, if they are good, say so If they are bad, say so (and say why) If there are issues with my idea, say what they are.
Ros-dev mailing list Ros-dev@reactos.com http://reactos.com:8080/mailman/listinfo/ros-dev
This sounds good. What will happen to the msvcrt headers? msvcrt itself (and things like crtdll and the others that build off msvcrt) should put all their internal stuff (including internal copies of the public headers like Microsoft does if that is required) But I think that any part of ReactOS that needs to talk to msvcrt/crtdll/etc should use mingw-runtime (i.e. we make it better using stuff in our msvcrt headers if we need to then use and import if needed mingw-runtime)
as for the other stuff, I assume that /sdk will basicly be w32api and that /ddk will be the w32api DDK. Anything that w32api wont take because its documented but not publically (e.g. the aformentioned fdi.h and fci.h) should go into the NDK IMO. I assume then that ndk will contain all the stuff that is exported by the various parts of windows but is not documented outside of microsoft. Assuming this to be the case, I suggest it is split up (logically if not physically split up into different bits) into a set of kernel headers (basicly everything exported by ntoskrnl, hal and the various kernel mode drivers like win32k.sys, videoprt.sys, ndis.sys etc) and a set of userland headers (i.e. all the stuff exported from userland dlls that microsoft doesnt document). Also, the NDK should contain something saying that this stuff is undocumented and there are no gaurantees that, if you use it, your program will work in the future or that it will work on real windows if microsoft changes the undocumented stuff)
IMO, anything that is exported by a dll or driver or whatever in ReactOS but which is not exported by the MS dlls should be seperate from the NDK (at least in a seperate set of headers anyway) and it should be clear that using these functions prevents your code from running on real windows.
Jonathan Wilson wrote:
This sounds good. What will happen to the msvcrt headers? msvcrt itself (and things like crtdll and the others that build off msvcrt) should put all their internal stuff (including internal copies of the public headers like Microsoft does if that is required) But I think that any part of ReactOS that needs to talk to msvcrt/crtdll/etc should use mingw-runtime (i.e. we make it better using stuff in our msvcrt headers if we need to then use and import if needed mingw-runtime)
I don't have much experience with those headers... I always assumed they were provided by deafault by mingw+gcc.
as for the other stuff, I assume that /sdk will basicly be w32api
Yes, with documented additions.
and that /ddk will be the w32api DDK.
As well as the IFS, with documented additions that we have in our "internal" ddk/ifs
Anything that w32api wont take because its documented but not publically (e.g. the aformentioned fdi.h and fci.h) should go into the NDK IMO.
Yes, the NDK has undocumented windows apis organized much like our old headers: "kefuncs.h/ketypes.h" "exfuncs.h/extypes.h" as well as "umtypes.h" which contains User-Mode Native types.
I assume then that ndk will contain all the stuff that is exported by the various parts of windows but is not documented outside of microsoft.
Yes.
Assuming this to be the case, I suggest it is split up (logically if not physically split up into different bits) into a set of kernel headers (basicly everything exported by ntoskrnl,
See the funcs/types organization.
hal
Halfuncs/Types
and the various kernel mode drivers like win32k.sys
I haven't thought about exported win32k functions yet... I haven't met any yet. But I suppose w32kfuncs/w32ktypes will do, but I'm still not sure if those should be in the NDK...I've never heard about many people calling them.
, videoprt.sys, ndis.sys etc)
Once again, I'm not sure those have exported functions that are usually being called, so I don't see why to include them in the NDK. They should just have their own set of internal headers.
and a set of userland headers (i.e. all the stuff exported from userland dlls that microsoft doesnt document).
Once again, the NDK is mostly for kernel functions that are either native or undocumented...so undocumented userland dlls (apart ntdll) probably don't have much use being called. But if it's something really useful, then they'll be documented.
Also, the NDK should contain something saying that this stuff is undocumented and there are no gaurantees that, if you use it, your program will work in the future or that it will work on real windows if microsoft changes the undocumented stuff)
It's the whole point of a NDK :)
IMO, anything that is exported by a dll or driver or whatever in ReactOS but which is not exported by the MS dlls should be seperate from the NDK (at least in a seperate set of headers anyway)
Of course. For now there's a rosfuncs/types in inclue/reactos in my new design.
and it should be clear that using these functions prevents your code from running on real windows.
Yes, the best would be to declare them as deprecated or somethign to make sure that nobody can use them in a compiled app/driver.
Best regards, Alex Ionescu
, videoprt.sys, ndis.sys etc)
Once again, I'm not sure those have exported functions that are usually being called, so I don't see why to include them in the NDK. They should just have their own set of internal headers.
just ask vizzini (our ndis guru) about ndis.sys and calling functions in it that are not documented by microsoft. I suspect that there are things in videoprt.sys not documented by microsoft either. IMO the NDK should basicly be 3 things: 1.a "native" API header file or files (i.e. the "nt.h" which has been referenced in at least one microsoft example or code somewhere) and would provide interfaces for everything in ntdll.dll that we can get prototypes for. 2.a set of header files that document everything exported by ntoskrnl, hal, win32k (the Engxxx functions), videoprt.sys, ndis.sys, scsiport.sys etc and not documented by microsoft. then 3.anything else in userland that isnt documeneted (e.g. undocumented shell32 functions or undocumented user32 functions or whatever else). I have actually seen some of this undocumented stuff in real-world usage. Lots of people use undocumented shell32 functions for various things. And several code examples out there subclass the menu windows in that app with a hook and hook various undocumented menu messages (I have seen references to MN_xxx messages that I think are the messages in question although I cant remember where I saw them). So, people are using the undocumented functionality. Since we are implementing it (to be compatible with windows), I see no reason not to document it somewhere for all to see...
Are we going to actually get permission to re-licence then submit all the w23api changes back to w32api? If so, I suggest we keep the stuff (like IFS) that w32api wont accept seperate (i.e. in seperate header files) from the base w32api (i.e. our w32api would only be w32api + stuff we are working on getting into w32api)
Given that anything that goes into our SDK and DDK headers will be stuff that is documented in the platform SDK with any ros-specific additions (e.g. for stuff that is documented)
--- Jonathan Wilson jonwil@tpgi.com.au wrote:
Are we going to actually get permission to re-licence then submit all the w23api changes back to w32api?
Unless its its on MSDN or something thats undocumented that you can find via a simple google search then the w32api people will not accept it. As far as I care we should just maintain our own SDK based on the Wine headers, The DDK based on the Mingw DDK and our own NDK for the undocumented stuff.
Thanks Steven
__________________________________ Do you Yahoo!? The all-new My Yahoo! - Get yours free! http://my.yahoo.com
Steven Edwards wrote:
--- Jonathan Wilson jonwil@tpgi.com.au wrote:
Are we going to actually get permission to re-licence then submit all the w23api changes back to w32api?
Unless its its on MSDN or something thats undocumented that you can find via a simple google search then the w32api people will not accept it. As far as I care we should just maintain our own SDK based on the Wine headers, The DDK based on the Mingw DDK and our own NDK for the undocumented stuff.
Most of what is in the WINE headers is the stuff that is in the "platform SDK" and then "the undocumented stuff". Keeping the "undocumented" stuff (i.e. the bits microsoft doesnt document in the Platform SDK) and the "documented" stuff seperate (i.e. putting the undocumented stuff into seperate headers) is a good idea IMO. (e.g. have a "userundoc.h" for the undocumented stuff or whatever) And if we are doing that, we can (if people will re-licence) contribute the documented stuff (all of which is in the MS platform SDK or the DDK) back to w32api itself to make w32api into something that gets 1 step closer to the Platform SDK/DDK (IMO thats what w32api should be aiming at, a clone of anything in the microsoft Platform SDK that someone wants to cleanroom clone and also anything in the DDK that anyone wants to cleanroom clone)
Anyone else think that keeping the undocumented stuff seperate from the regular documented stuff is a good idea? bad idea?
Jonathan Wilson wrote:
Steven Edwards wrote:
--- Jonathan Wilson jonwil@tpgi.com.au wrote:
Are we going to actually get permission to re-licence then submit all the w23api changes back to w32api?
Unless its its on MSDN or something thats undocumented that you can find via a simple google search then the w32api people will not accept it. As far as I care we should just maintain our own SDK based on the Wine headers, The DDK based on the Mingw DDK and our own NDK for the undocumented stuff.
Most of what is in the WINE headers is the stuff that is in the "platform SDK" and then "the undocumented stuff". Keeping the "undocumented" stuff (i.e. the bits microsoft doesnt document in the Platform SDK) and the "documented" stuff seperate (i.e. putting the undocumented stuff into seperate headers) is a good idea IMO. (e.g. have a "userundoc.h" for the undocumented stuff or whatever) And if we are doing that, we can (if people will re-licence) contribute the documented stuff (all of which is in the MS platform SDK or the DDK) back to w32api itself to make w32api into something that gets 1 step closer to the Platform SDK/DDK (IMO thats what w32api should be aiming at, a clone of anything in the microsoft Platform SDK that someone wants to cleanroom clone and also anything in the DDK that anyone wants to cleanroom clone)
Anyone else think that keeping the undocumented stuff seperate from the regular documented stuff is a good idea? bad idea?
We will do as planned and put undocumented stuff in the NDK
Best regards, Alex Ionescu
-----Original Message----- From: ros-dev-bounces@reactos.com [mailto:ros-dev-bounces@reactos.com] On Behalf Of Steven Edwards Sent: 21. november 2004 01:43 To: ReactOS Development List Subject: Re: [ros-dev] headers suggestion
--- Jonathan Wilson jonwil@tpgi.com.au wrote:
Are we going to actually get permission to re-licence then
submit all
the w23api changes back to w32api?
Unless its its on MSDN or something thats undocumented that you can find via a simple google search then the w32api people will not accept it. As far as I care we should just maintain our own SDK based on the Wine headers, The DDK based on the Mingw DDK and our own NDK for the undocumented stuff.
Thanks Steven
I don't understand why you want two free SDK projects.
Casper
--- Casper Hornstrup chorns@users.sourceforge.net wrote:
I don't understand why you want two free SDK projects.
I am not wanting to create two free SDK Projects. Lets be clear the the mingw SDK/DDK is good but the rules for getting anything in to it are crazy. MSDN has been known to be wrong almost every other day so using it as a primary source of information flawed. The Wine headers already exist so its not like I am talking about creating something new here. They have shown a willingness to work with us so we should work with them.
Thanks Steven
__________________________________ Do you Yahoo!? The all-new My Yahoo! - Get yours free! http://my.yahoo.com
-----Original Message----- From: ros-dev-bounces@reactos.com [mailto:ros-dev-bounces@reactos.com] On Behalf Of Steven Edwards Sent: 21. november 2004 15:03 To: ReactOS Development List Subject: RE: [ros-dev] headers suggestion
--- Casper Hornstrup chorns@users.sourceforge.net wrote:
I don't understand why you want two free SDK projects.
I am not wanting to create two free SDK Projects. Lets be clear the the mingw SDK/DDK is good but the rules for getting anything in to it are crazy. MSDN has been known to be wrong almost every other day so using it as a primary source of information flawed. The Wine headers already exist so its not like I am talking about creating something new here. They have shown a willingness to work with us so we should work with them.
Thanks Steven
Which rules are crazy? MSDN is usually right btw. You are greatly exaggerating. You have to see it in the bigger perspective. MSDN has tons of information and is made by humans and since humans tend to make mistakes every now and then, MSDN will have them too. Why don't you post in their newsgroups about the mistakes you've found so they have a chance of fixing them? I've found the MS newgroups to be a valuable way to get your voice heard and to get help with problems.
Casper
Casper,
--- Casper Hornstrup chorns@users.sourceforge.net wrote:
Which rules are crazy? MSDN is usually right btw. You are greatly exaggerating. You have to see it in the bigger perspective. MSDN has tons of information and is made by humans and since humans tend to make mistakes every now and then, MSDN will have them too. Why don't you post in their newsgroups about the mistakes you've found so they have a chance of fixing them? I've found the MS newgroups to be a valuable way to get your voice heard and to get help with problems.
Maybe I should have been more clear. Unless its on MSDN and does not require downloading a SDK package then they will accept it. Never mind the fact that the cabinet.dll information is published in a SDK on MSDN and the PSDK. They will not accept it as you have to click though a EULA to get to it. I know. I went through the trouble of asking the Wine authors to relicense the FCI/FDI headers only to have the Mingw people say "find it on MSDN" which I did, in a downloadable SDK. After I sent that they said "we can not accept this, it has a EULA"
Take a look at the headers that are in Wine that are missing in our w32api. I doubt most of them are even in the latest w32api. The Wine people are happy to work with us to fix them. Should we rewrite them from scratch only to be told by the Mingw people sorry MSDN has a broken URL or what you submit does not match, even if I can prove it is wrong via a real world regression test.
I don't want to fight about it. If you want to duplicate a lot of work go ahead and I will keep my mouth shut.
Thanks Steven
__________________________________ Do you Yahoo!? Meet the all-new My Yahoo! - Try it today! http://my.yahoo.com
Steven Edwards wrote:
Casper,
--- Casper Hornstrup chorns@users.sourceforge.net wrote:
Which rules are crazy? MSDN is usually right btw. You are greatly exaggerating. You have to see it in the bigger perspective. MSDN has tons of information and is made by humans and since humans tend to make mistakes every now and then, MSDN will have them too. Why don't you post in their newsgroups about the mistakes you've found so they have a chance of fixing them? I've found the MS newgroups to be a valuable way to get your voice heard and to get help with problems.
Maybe I should have been more clear. Unless its on MSDN and does not require downloading a SDK package then they will accept it. Never mind the fact that the cabinet.dll information is published in a SDK on MSDN and the PSDK. They will not accept it as you have to click though a EULA to get to it. I know. I went through the trouble of asking the Wine authors to relicense the FCI/FDI headers only to have the Mingw people say "find it on MSDN" which I did, in a downloadable SDK. After I sent that they said "we can not accept this, it has a EULA"
Take a look at the headers that are in Wine that are missing in our w32api. I doubt most of them are even in the latest w32api. The Wine people are happy to work with us to fix them. Should we rewrite them from scratch only to be told by the Mingw people sorry MSDN has a broken URL or what you submit does not match, even if I can prove it is wrong via a real world regression test.
I don't want to fight about it. If you want to duplicate a lot of work go ahead and I will keep my mouth shut.
Thanks Steven
__________________________________ Do you Yahoo!? Meet the all-new My Yahoo! - Try it today! http://my.yahoo.com
Ros-dev mailing list Ros-dev@reactos.com http://reactos.com:8080/mailman/listinfo/ros-dev
Let's all just wait for what I'm working on and then you can all flame me instead of arguing with each others :)
Best regards, Alex Ionescu
-
Alex Ionescu -
Casper Hornstrup -
Jonathan Wilson -
Steven Edwards